CVE-2026-7141: Vllm Kv Cache Uninitialized Resource
KV cache blocks for FullAttention models are not zeroed due to incomplete logic in needs_kv_cache_zeroing.
Provally CuratedPublic repositoryHighMedium confidenceVerifiedApache-2.0Python
greprules fetch cve-2026-7141-vllm-kv-cache-uninitialized-resource --engine opengrepDescription
KV cache blocks for FullAttention models are not zeroed due to incomplete logic in needs_kv_cache_zeroing.
Community feedback
0 signals from signed-in users.
- Useful
- 0
- False positive
- 0
- Metadata
- 0